ADMT SDK

Prove your AI is advisory-only.

Cryptographic audit trails for ADMT compliance. Three-prong proof that humans make the decisions.

Coming Soon

ONE PACKAGE, THREE PRODUCTS

Advisor

Ships first

Proves AI is advisory-only via cryptographic audit trail. Write isolation, dwell time enforcement, and rubber-stamp detection.

Write isolationDwell time enforcementThree-prong proofRubber-stamp detection

Auditor

Generate §7155 assessment documents automatically. Interview flow, consumer request handling, and lifecycle management.

Interview flowAssessment generationConsumer requestsLifecycle management

Trainer

Decision corpus analytics and automation readiness scoring. Understand when humans agree with AI — and when they don't.

Agreement rate analyticsReadiness scoringCustom model transitionDecision corpus builder

QUICKSTART

Four steps to a cryptographic audit trail.

STEP 01

Configure

Set up storage backend, dwell time minimums, and enforcement mode. The gateway sits between your AI and your decision layer.

python
from admt import AdvisorConfig, Gateway

config = AdvisorConfig(
    storage_backend="sqlite",
    dwell_time_seconds=30.0,
    enforcement="advisory",
)

gateway = Gateway(config)
STEP 02

Recommend

AI submits a recommendation through the gateway. Write isolation ensures it can never modify the final decision directly.

python
from admt import DecisionInput

rec = await gateway.recommend(
    DecisionInput(
        action="approve_loan",
        subject_id="applicant-4821",
        confidence=0.92,
        factors=["credit_score", "income_ratio"],
    )
)

# rec.id → track through review
STEP 03

Review & Decide

Human reviews the recommendation, engages with the evidence, and submits an independent decision. Dwell time is enforced automatically.

python
decision = await gateway.decide(
    recommendation_id=rec.id,
    reviewer_id="analyst-jane",
    action="approve_loan",
    rationale="Strong credit history, DTI within policy",
    factors_reviewed=["credit_score", "income_ratio"],
)

# Dwell time enforced — rejects if too fast
STEP 04

Verify

Verify hash chain integrity and generate three-prong proof. Every event is Ed25519 signed and SHA-256 chained — tamper detection is built in.

python
proof = await gateway.verify(decision.id)

assert proof.chain_valid        # SHA-256 hash chain intact
assert proof.signature_valid    # Ed25519 signatures verified
assert proof.three_prong.met    # All three prongs satisfied

# Export for regulators
report = proof.export(format="pdf")

BUILT FOR COMPLIANCE

Cryptographic Audit Trail

Ed25519 signed, SHA-256 hash-chained, append-only event log. Every recommendation, review, and decision is immutably recorded.

Three-Prong Proof

Automated evidence of meaningful human review: recommendation presented, reviewer engaged, decision reflects independent judgment.

Dwell Time Enforcement

Configurable minimum review times. Detects rubber-stamping patterns and flags insufficient human engagement.

Write Isolation

AI architecturally cannot modify decisions. Recommendations flow through a read-only gateway — humans hold the pen.

Deep Review Sampling

Random audit of decision quality. Surfaces patterns in human overrides and agreement rates across decision categories.

Framework Ready

FastAPI router, LangGraph integration, CrewAI hooks. Drop into your existing AI stack without rewiring anything.

Get notified when we launch.

Join the waitlist for early access to the ADMT SDK.